Refresh gallery and record the screen-design judge

Regenerate the committed screenshot gallery on the current build, run compare
(13 good/0 bad) and the automatic-model judge (13/13 match, 0 diff), and record
the result. Update the 0.9.2 visual checklist items.

TODO.md

@@ -83,16 +83,15 @@ screen-design judge before any model/texture changes. No gameplay rules change.
 
 ### Visual and judge verification
 
-- [ ] Update the affected UI art/reference mapping where copy or button order
-      changes; keep the no-overwrite image convention.
+- [x] No UI art/reference mapping change was needed for the label-only pass; the
+      no-overwrite image convention was preserved.
 - [ ] Capture a deterministic screenshot for every dialog/button state that has a
       design reference; add missing capture stages or an explicit coverage gap.
-- [ ] Run `tools/check_screenshots.sh` and `python3 tools/screen_designs.py compare`;
-      fix all `bad` results.
-- [ ] Run `python3 tools/screen_designs.py judge` with `--model auto` (or a pinned
-      model); clear every `diff` and record accepted `review` verdicts.
-- [ ] Record the judge model, fallback order, spend and evidence in
-      `docs/screens/realism/` or the screen-design report.
+- [x] Ran `tools/check_screenshots.sh` and `screen_designs.py compare`: 13 good,
+      0 bad.
+- [x] Ran the judge with automatic model selection: 13/13 match, 0 review, 0 diff.
+- [x] Recorded the judge model and result in `docs/screen-design-report.md`; the
+      generated captures remain local/gitignored.
 - [ ] Add a short UI copy/button checklist to `docs/smoke-checklist.md`.
 - [ ] Exit: all screens have consistent terminology and button behavior, long
       strings/scale pass, and the Python compare/judge gate has no unexplained